net/ipv4/ipvs/ip_vs_conn.c

   1 /*
   2  * IPVS         An implementation of the IP virtual server support for the
   3  *              LINUX operating system.  IPVS is now implemented as a module
   4  *              over the Netfilter framework. IPVS can be used to build a
   5  *              high-performance and highly available server based on a
   6  *              cluster of servers.
   7  *
   8  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   9  *              Peter Kese <peter.kese@ijs.si>
  10  *              Julian Anastasov <ja@ssi.bg>
  11  *
  12  *              This program is free software; you can redistribute it and/or
  13  *              modify it under the terms of the GNU General Public License
  14  *              as published by the Free Software Foundation; either version
  15  *              2 of the License, or (at your option) any later version.
  16  *
  17  * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
  18  * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
  19  * and others. Many code here is taken from IP MASQ code of kernel 2.2.
  20  *
  21  * Changes:
  22  *
  23  */
  24
  25 #include <linux/interrupt.h>
  26 #include <linux/in.h>
  27 #include <linux/net.h>
  28 #include <linux/kernel.h>
  29 #include <linux/module.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/proc_fs.h>              /* for proc_net_* */
  32 #include <linux/seq_file.h>
  33 #include <linux/jhash.h>
  34 #include <linux/random.h>
  35
  36 #include <net/net_namespace.h>
  37 #include <net/ip_vs.h>
  38
  39
  40 /*
  41  *  Connection hash table: for input and output packets lookups of IPVS
  42  */
  43 static struct list_head *ip_vs_conn_tab;
  44
  45 /*  SLAB cache for IPVS connections */
  46 static struct kmem_cache *ip_vs_conn_cachep __read_mostly;
  47
  48 /*  counter for current IPVS connections */
  49 static atomic_t ip_vs_conn_count = ATOMIC_INIT(0);
  50
  51 /*  counter for no client port connections */
  52 static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0);
  53
  54 /* random value for IPVS connection hash */
  55 static unsigned int ip_vs_conn_rnd;
  56
  57 /*
  58  *  Fine locking granularity for big connection hash table
  59  */
  60 #define CT_LOCKARRAY_BITS  4
  61 #define CT_LOCKARRAY_SIZE  (1<<CT_LOCKARRAY_BITS)
  62 #define CT_LOCKARRAY_MASK  (CT_LOCKARRAY_SIZE-1)
  63
  64 struct ip_vs_aligned_lock
  65 {
  66         rwlock_t        l;
  67 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
  68
  69 /* lock array for conn table */
  70 static struct ip_vs_aligned_lock
  71 __ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
  72
  73 static inline void ct_read_lock(unsigned key)
  74 {
  75         read_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
  76 }
  77
  78 static inline void ct_read_unlock(unsigned key)
  79 {
  80         read_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
  81 }
  82
  83 static inline void ct_write_lock(unsigned key)
  84 {
  85         write_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
  86 }
  87
  88 static inline void ct_write_unlock(unsigned key)
  89 {
  90         write_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
  91 }
  92
  93 static inline void ct_read_lock_bh(unsigned key)
  94 {
  95         read_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
  96 }
  97
  98 static inline void ct_read_unlock_bh(unsigned key)
  99 {
 100         read_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 101 }
 102
 103 static inline void ct_write_lock_bh(unsigned key)
 104 {
 105         write_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 106 }
 107
 108 static inline void ct_write_unlock_bh(unsigned key)
 109 {
 110         write_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 111 }
 112
 113
 114 /*
 115  *      Returns hash value for IPVS connection entry
 116  */
 117 static unsigned int ip_vs_conn_hashkey(int af, unsigned proto,
 118                                        const union nf_inet_addr *addr,
 119                                        __be16 port)
 120 {
 121 #ifdef CONFIG_IP_VS_IPV6
 122         if (af == AF_INET6)
 123                 return jhash_3words(jhash(addr, 16, ip_vs_conn_rnd),
 124                                     (__force u32)port, proto, ip_vs_conn_rnd)
 125                         & IP_VS_CONN_TAB_MASK;
 126 #endif
 127         return jhash_3words((__force u32)addr->ip, (__force u32)port, proto,
 128                             ip_vs_conn_rnd)
 129                 & IP_VS_CONN_TAB_MASK;
 130 }
 131
 132
 133 /*
 134  *      Hashes ip_vs_conn in ip_vs_conn_tab by proto,addr,port.
 135  *      returns bool success.
 136  */
 137 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
 138 {
 139         unsigned hash;
 140         int ret;
 141
 142         /* Hash by protocol, client address and port */
 143         hash = ip_vs_conn_hashkey(cp->af, cp->protocol, &cp->caddr, cp->cport);
 144
 145         ct_write_lock(hash);
 146
 147         if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
 148                 list_add(&cp->c_list, &ip_vs_conn_tab[hash]);
 149                 cp->flags |= IP_VS_CONN_F_HASHED;
 150                 atomic_inc(&cp->refcnt);
 151                 ret = 1;
 152         } else {
 153                 IP_VS_ERR("ip_vs_conn_hash(): request for already hashed, "
 154                           "called from %p\n", __builtin_return_address(0));
 155                 ret = 0;
 156         }
 157
 158         ct_write_unlock(hash);
 159
 160         return ret;
 161 }
 162
 163
 164 /*
 165  *      UNhashes ip_vs_conn from ip_vs_conn_tab.
 166  *      returns bool success.
 167  */
 168 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
 169 {
 170         unsigned hash;
 171         int ret;
 172
 173         /* unhash it and decrease its reference counter */
 174         hash = ip_vs_conn_hashkey(cp->af, cp->protocol, &cp->caddr, cp->cport);
 175
 176         ct_write_lock(hash);
 177
 178         if (cp->flags & IP_VS_CONN_F_HASHED) {
 179                 list_del(&cp->c_list);
 180                 cp->flags &= ~IP_VS_CONN_F_HASHED;
 181                 atomic_dec(&cp->refcnt);
 182                 ret = 1;
 183         } else
 184                 ret = 0;
 185
 186         ct_write_unlock(hash);
 187
 188         return ret;
 189 }
 190
 191
 192 /*
 193  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
 194  *  Called for pkts coming from OUTside-to-INside.
 195  *      s_addr, s_port: pkt source address (foreign host)
 196  *      d_addr, d_port: pkt dest address (load balancer)
 197  */
 198 static inline struct ip_vs_conn *__ip_vs_conn_in_get
 199 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
 200  const union nf_inet_addr *d_addr, __be16 d_port)
 201 {
 202         unsigned hash;
 203         struct ip_vs_conn *cp;
 204
 205         hash = ip_vs_conn_hashkey(af, protocol, s_addr, s_port);
 206
 207         ct_read_lock(hash);
 208
 209         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
 210                 if (cp->af == af &&
 211                     ip_vs_addr_equal(af, s_addr, &cp->caddr) &&
 212                     ip_vs_addr_equal(af, d_addr, &cp->vaddr) &&
 213                     s_port == cp->cport && d_port == cp->vport &&
 214                     ((!s_port) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
 215                     protocol == cp->protocol) {
 216                         /* HIT */
 217                         atomic_inc(&cp->refcnt);
 218                         ct_read_unlock(hash);
 219                         return cp;
 220                 }
 221         }
 222
 223         ct_read_unlock(hash);
 224
 225         return NULL;
 226 }
 227
 228 struct ip_vs_conn *ip_vs_conn_in_get
 229 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
 230  const union nf_inet_addr *d_addr, __be16 d_port)
 231 {
 232         struct ip_vs_conn *cp;
 233
 234         cp = __ip_vs_conn_in_get(af, protocol, s_addr, s_port, d_addr, d_port);
 235         if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt))
 236                 cp = __ip_vs_conn_in_get(af, protocol, s_addr, 0, d_addr,
 237                                          d_port);
 238
 239         IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
 240                       ip_vs_proto_name(protocol),
 241                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
 242                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
 243                       cp ? "hit" : "not hit");
 244
 245         return cp;
 246 }
 247
 248 /* Get reference to connection template */
 249 struct ip_vs_conn *ip_vs_ct_in_get
 250 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
 251  const union nf_inet_addr *d_addr, __be16 d_port)
 252 {
 253         unsigned hash;
 254         struct ip_vs_conn *cp;
 255
 256         hash = ip_vs_conn_hashkey(af, protocol, s_addr, s_port);
 257
 258         ct_read_lock(hash);
 259
 260         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
 261                 if (cp->af == af &&
 262                     ip_vs_addr_equal(af, s_addr, &cp->caddr) &&
 263                     ip_vs_addr_equal(af, d_addr, &cp->vaddr) &&
 264                     s_port == cp->cport && d_port == cp->vport &&
 265                     cp->flags & IP_VS_CONN_F_TEMPLATE &&
 266                     protocol == cp->protocol) {
 267                         /* HIT */
 268                         atomic_inc(&cp->refcnt);
 269                         goto out;
 270                 }
 271         }
 272         cp = NULL;
 273
 274   out:
 275         ct_read_unlock(hash);
 276
 277         IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
 278                       ip_vs_proto_name(protocol),
 279                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
 280                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
 281                       cp ? "hit" : "not hit");
 282
 283         return cp;
 284 }
 285
 286 /*
 287  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
 288  *  Called for pkts coming from inside-to-OUTside.
 289  *      s_addr, s_port: pkt source address (inside host)
 290  *      d_addr, d_port: pkt dest address (foreign host)
 291  */
 292 struct ip_vs_conn *ip_vs_conn_out_get
 293 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
 294  const union nf_inet_addr *d_addr, __be16 d_port)
 295 {
 296         unsigned hash;
 297         struct ip_vs_conn *cp, *ret=NULL;
 298
 299         /*
 300          *      Check for "full" addressed entries
 301          */
 302         hash = ip_vs_conn_hashkey(af, protocol, d_addr, d_port);
 303
 304         ct_read_lock(hash);
 305
 306         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
 307                 if (cp->af == af &&
 308                     ip_vs_addr_equal(af, d_addr, &cp->caddr) &&
 309                     ip_vs_addr_equal(af, s_addr, &cp->daddr) &&
 310                     d_port == cp->cport && s_port == cp->dport &&
 311                     protocol == cp->protocol) {
 312                         /* HIT */
 313                         atomic_inc(&cp->refcnt);
 314                         ret = cp;
 315                         break;
 316                 }
 317         }
 318
 319         ct_read_unlock(hash);
 320
 321         IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
 322                       ip_vs_proto_name(protocol),
 323                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
 324                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
 325                       ret ? "hit" : "not hit");
 326
 327         return ret;
 328 }
 329
 330
 331 /*
 332  *      Put back the conn and restart its timer with its timeout
 333  */
 334 void ip_vs_conn_put(struct ip_vs_conn *cp)
 335 {
 336         /* reset it expire in its timeout */
 337         mod_timer(&cp->timer, jiffies+cp->timeout);
 338
 339         __ip_vs_conn_put(cp);
 340 }
 341
 342
 343 /*
 344  *      Fill a no_client_port connection with a client port number
 345  */
 346 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
 347 {
 348         if (ip_vs_conn_unhash(cp)) {
 349                 spin_lock(&cp->lock);
 350                 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
 351                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 352                         cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
 353                         cp->cport = cport;
 354                 }
 355                 spin_unlock(&cp->lock);
 356
 357                 /* hash on new dport */
 358                 ip_vs_conn_hash(cp);
 359         }
 360 }
 361
 362
 363 /*
 364  *      Bind a connection entry with the corresponding packet_xmit.
 365  *      Called by ip_vs_conn_new.
 366  */
 367 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
 368 {
 369         switch (IP_VS_FWD_METHOD(cp)) {
 370         case IP_VS_CONN_F_MASQ:
 371                 cp->packet_xmit = ip_vs_nat_xmit;
 372                 break;
 373
 374         case IP_VS_CONN_F_TUNNEL:
 375                 cp->packet_xmit = ip_vs_tunnel_xmit;
 376                 break;
 377
 378         case IP_VS_CONN_F_DROUTE:
 379                 cp->packet_xmit = ip_vs_dr_xmit;
 380                 break;
 381
 382         case IP_VS_CONN_F_LOCALNODE:
 383                 cp->packet_xmit = ip_vs_null_xmit;
 384                 break;
 385
 386         case IP_VS_CONN_F_BYPASS:
 387                 cp->packet_xmit = ip_vs_bypass_xmit;
 388                 break;
 389         }
 390 }
 391
 392 #ifdef CONFIG_IP_VS_IPV6
 393 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
 394 {
 395         switch (IP_VS_FWD_METHOD(cp)) {
 396         case IP_VS_CONN_F_MASQ:
 397                 cp->packet_xmit = ip_vs_nat_xmit_v6;
 398                 break;
 399
 400         case IP_VS_CONN_F_TUNNEL:
 401                 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
 402                 break;
 403
 404         case IP_VS_CONN_F_DROUTE:
 405                 cp->packet_xmit = ip_vs_dr_xmit_v6;
 406                 break;
 407
 408         case IP_VS_CONN_F_LOCALNODE:
 409                 cp->packet_xmit = ip_vs_null_xmit;
 410                 break;
 411
 412         case IP_VS_CONN_F_BYPASS:
 413                 cp->packet_xmit = ip_vs_bypass_xmit_v6;
 414                 break;
 415         }
 416 }
 417 #endif
 418
 419
 420 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
 421 {
 422         return atomic_read(&dest->activeconns)
 423                 + atomic_read(&dest->inactconns);
 424 }
 425
 426 /*
 427  *      Bind a connection entry with a virtual service destination
 428  *      Called just after a new connection entry is created.
 429  */
 430 static inline void
 431 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
 432 {
 433         /* if dest is NULL, then return directly */
 434         if (!dest)
 435                 return;
 436
 437         /* Increase the refcnt counter of the dest */
 438         atomic_inc(&dest->refcnt);
 439
 440         /* Bind with the destination and its corresponding transmitter */
 441         if ((cp->flags & IP_VS_CONN_F_SYNC) &&
 442             (!(cp->flags & IP_VS_CONN_F_TEMPLATE)))
 443                 /* if the connection is not template and is created
 444                  * by sync, preserve the activity flag.
 445                  */
 446                 cp->flags |= atomic_read(&dest->conn_flags) &
 447                              (~IP_VS_CONN_F_INACTIVE);
 448         else
 449                 cp->flags |= atomic_read(&dest->conn_flags);
 450         cp->dest = dest;
 451
 452         IP_VS_DBG(7, "Bind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
 453                   "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 454                   "dest->refcnt:%d\n",
 455                   ip_vs_proto_name(cp->protocol),
 456                   NIPQUAD(cp->caddr.ip), ntohs(cp->cport),
 457                   NIPQUAD(cp->vaddr.ip), ntohs(cp->vport),
 458                   NIPQUAD(cp->daddr.ip), ntohs(cp->dport),
 459                   ip_vs_fwd_tag(cp), cp->state,
 460                   cp->flags, atomic_read(&cp->refcnt),
 461                   atomic_read(&dest->refcnt));
 462
 463         /* Update the connection counters */
 464         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
 465                 /* It is a normal connection, so increase the inactive
 466                    connection counter because it is in TCP SYNRECV
 467                    state (inactive) or other protocol inacive state */
 468                 if ((cp->flags & IP_VS_CONN_F_SYNC) &&
 469                     (!(cp->flags & IP_VS_CONN_F_INACTIVE)))
 470                         atomic_inc(&dest->activeconns);
 471                 else
 472                         atomic_inc(&dest->inactconns);
 473         } else {
 474                 /* It is a persistent connection/template, so increase
 475                    the peristent connection counter */
 476                 atomic_inc(&dest->persistconns);
 477         }
 478
 479         if (dest->u_threshold != 0 &&
 480             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
 481                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
 482 }
 483
 484
 485 /*
 486  * Check if there is a destination for the connection, if so
 487  * bind the connection to the destination.
 488  */
 489 struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
 490 {
 491         struct ip_vs_dest *dest;
 492
 493         if ((cp) && (!cp->dest)) {
 494                 dest = ip_vs_find_dest(cp->af, &cp->daddr, cp->dport,
 495                                        &cp->vaddr, cp->vport,
 496                                        cp->protocol);
 497                 ip_vs_bind_dest(cp, dest);
 498                 return dest;
 499         } else
 500                 return NULL;
 501 }
 502
 503
 504 /*
 505  *      Unbind a connection entry with its VS destination
 506  *      Called by the ip_vs_conn_expire function.
 507  */
 508 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
 509 {
 510         struct ip_vs_dest *dest = cp->dest;
 511
 512         if (!dest)
 513                 return;
 514
 515         IP_VS_DBG(7, "Unbind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
 516                   "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 517                   "dest->refcnt:%d\n",
 518                   ip_vs_proto_name(cp->protocol),
 519                   NIPQUAD(cp->caddr.ip), ntohs(cp->cport),
 520                   NIPQUAD(cp->vaddr.ip), ntohs(cp->vport),
 521                   NIPQUAD(cp->daddr.ip), ntohs(cp->dport),
 522                   ip_vs_fwd_tag(cp), cp->state,
 523                   cp->flags, atomic_read(&cp->refcnt),
 524                   atomic_read(&dest->refcnt));
 525
 526         /* Update the connection counters */
 527         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
 528                 /* It is a normal connection, so decrease the inactconns
 529                    or activeconns counter */
 530                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
 531                         atomic_dec(&dest->inactconns);
 532                 } else {
 533                         atomic_dec(&dest->activeconns);
 534                 }
 535         } else {
 536                 /* It is a persistent connection/template, so decrease
 537                    the peristent connection counter */
 538                 atomic_dec(&dest->persistconns);
 539         }
 540
 541         if (dest->l_threshold != 0) {
 542                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
 543                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 544         } else if (dest->u_threshold != 0) {
 545                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
 546                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 547         } else {
 548                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 549                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 550         }
 551
 552         /*
 553          * Simply decrease the refcnt of the dest, because the
 554          * dest will be either in service's destination list
 555          * or in the trash.
 556          */
 557         atomic_dec(&dest->refcnt);
 558 }
 559
 560
 561 /*
 562  *      Checking if the destination of a connection template is available.
 563  *      If available, return 1, otherwise invalidate this connection
 564  *      template and return 0.
 565  */
 566 int ip_vs_check_template(struct ip_vs_conn *ct)
 567 {
 568         struct ip_vs_dest *dest = ct->dest;
 569
 570         /*
 571          * Checking the dest server status.
 572          */
 573         if ((dest == NULL) ||
 574             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
 575             (sysctl_ip_vs_expire_quiescent_template &&
 576              (atomic_read(&dest->weight) == 0))) {
 577                 IP_VS_DBG(9, "check_template: dest not available for "
 578                           "protocol %s s:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
 579                           "-> d:%u.%u.%u.%u:%d\n",
 580                           ip_vs_proto_name(ct->protocol),
 581                           NIPQUAD(ct->caddr.ip), ntohs(ct->cport),
 582                           NIPQUAD(ct->vaddr.ip), ntohs(ct->vport),
 583                           NIPQUAD(ct->daddr.ip), ntohs(ct->dport));
 584
 585                 /*
 586                  * Invalidate the connection template
 587                  */
 588                 if (ct->vport != htons(0xffff)) {
 589                         if (ip_vs_conn_unhash(ct)) {
 590                                 ct->dport = htons(0xffff);
 591                                 ct->vport = htons(0xffff);
 592                                 ct->cport = 0;
 593                                 ip_vs_conn_hash(ct);
 594                         }
 595                 }
 596
 597                 /*
 598                  * Simply decrease the refcnt of the template,
 599                  * don't restart its timer.
 600                  */
 601                 atomic_dec(&ct->refcnt);
 602                 return 0;
 603         }
 604         return 1;
 605 }
 606
 607 static void ip_vs_conn_expire(unsigned long data)
 608 {
 609         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
 610
 611         cp->timeout = 60*HZ;
 612
 613         /*
 614          *      hey, I'm using it
 615          */
 616         atomic_inc(&cp->refcnt);
 617
 618         /*
 619          *      do I control anybody?
 620          */
 621         if (atomic_read(&cp->n_control))
 622                 goto expire_later;
 623
 624         /*
 625          *      unhash it if it is hashed in the conn table
 626          */
 627         if (!ip_vs_conn_unhash(cp))
 628                 goto expire_later;
 629
 630         /*
 631          *      refcnt==1 implies I'm the only one referrer
 632          */
 633         if (likely(atomic_read(&cp->refcnt) == 1)) {
 634                 /* delete the timer if it is activated by other users */
 635                 if (timer_pending(&cp->timer))
 636                         del_timer(&cp->timer);
 637
 638                 /* does anybody control me? */
 639                 if (cp->control)
 640                         ip_vs_control_del(cp);
 641
 642                 if (unlikely(cp->app != NULL))
 643                         ip_vs_unbind_app(cp);
 644                 ip_vs_unbind_dest(cp);
 645                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
 646                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 647                 atomic_dec(&ip_vs_conn_count);
 648
 649                 kmem_cache_free(ip_vs_conn_cachep, cp);
 650                 return;
 651         }
 652
 653         /* hash it back to the table */
 654         ip_vs_conn_hash(cp);
 655
 656   expire_later:
 657         IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
 658                   atomic_read(&cp->refcnt)-1,
 659                   atomic_read(&cp->n_control));
 660
 661         ip_vs_conn_put(cp);
 662 }
 663
 664
 665 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
 666 {
 667         if (del_timer(&cp->timer))
 668                 mod_timer(&cp->timer, jiffies);
 669 }
 670
 671
 672 /*
 673  *      Create a new connection entry and hash it into the ip_vs_conn_tab
 674  */
 675 struct ip_vs_conn *
 676 ip_vs_conn_new(int af, int proto, const union nf_inet_addr *caddr, __be16 cport,
 677                const union nf_inet_addr *vaddr, __be16 vport,
 678                const union nf_inet_addr *daddr, __be16 dport, unsigned flags,
 679                struct ip_vs_dest *dest)
 680 {
 681         struct ip_vs_conn *cp;
 682         struct ip_vs_protocol *pp = ip_vs_proto_get(proto);
 683
 684         cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC);
 685         if (cp == NULL) {
 686                 IP_VS_ERR_RL("ip_vs_conn_new: no memory available.\n");
 687                 return NULL;
 688         }
 689
 690         INIT_LIST_HEAD(&cp->c_list);
 691         setup_timer(&cp->timer, ip_vs_conn_expire, (unsigned long)cp);
 692         cp->af             = af;
 693         cp->protocol       = proto;
 694         ip_vs_addr_copy(af, &cp->caddr, caddr);
 695         cp->cport          = cport;
 696         ip_vs_addr_copy(af, &cp->vaddr, vaddr);
 697         cp->vport          = vport;
 698         ip_vs_addr_copy(af, &cp->daddr, daddr);
 699         cp->dport          = dport;
 700         cp->flags          = flags;
 701         spin_lock_init(&cp->lock);
 702
 703         /*
 704          * Set the entry is referenced by the current thread before hashing
 705          * it in the table, so that other thread run ip_vs_random_dropentry
 706          * but cannot drop this entry.
 707          */
 708         atomic_set(&cp->refcnt, 1);
 709
 710         atomic_set(&cp->n_control, 0);
 711         atomic_set(&cp->in_pkts, 0);
 712
 713         atomic_inc(&ip_vs_conn_count);
 714         if (flags & IP_VS_CONN_F_NO_CPORT)
 715                 atomic_inc(&ip_vs_conn_no_cport_cnt);
 716
 717         /* Bind the connection with a destination server */
 718         ip_vs_bind_dest(cp, dest);
 719
 720         /* Set its state and timeout */
 721         cp->state = 0;
 722         cp->timeout = 3*HZ;
 723
 724         /* Bind its packet transmitter */
 725 #ifdef CONFIG_IP_VS_IPV6
 726         if (af == AF_INET6)
 727                 ip_vs_bind_xmit_v6(cp);
 728         else
 729 #endif
 730                 ip_vs_bind_xmit(cp);
 731
 732         if (unlikely(pp && atomic_read(&pp->appcnt)))
 733                 ip_vs_bind_app(cp, pp);
 734
 735         /* Hash it in the ip_vs_conn_tab finally */
 736         ip_vs_conn_hash(cp);
 737
 738         return cp;
 739 }
 740
 741
 742 /*
 743  *      /proc/net/ip_vs_conn entries
 744  */
 745 #ifdef CONFIG_PROC_FS
 746
 747 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
 748 {
 749         int idx;
 750         struct ip_vs_conn *cp;
 751
 752         for(idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) {
 753                 ct_read_lock_bh(idx);
 754                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
 755                         if (pos-- == 0) {
 756                                 seq->private = &ip_vs_conn_tab[idx];
 757                                 return cp;
 758                         }
 759                 }
 760                 ct_read_unlock_bh(idx);
 761         }
 762
 763         return NULL;
 764 }
 765
 766 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
 767 {
 768         seq->private = NULL;
 769         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
 770 }
 771
 772 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 773 {
 774         struct ip_vs_conn *cp = v;
 775         struct list_head *e, *l = seq->private;
 776         int idx;
 777
 778         ++*pos;
 779         if (v == SEQ_START_TOKEN)
 780                 return ip_vs_conn_array(seq, 0);
 781
 782         /* more on same hash chain? */
 783         if ((e = cp->c_list.next) != l)
 784                 return list_entry(e, struct ip_vs_conn, c_list);
 785
 786         idx = l - ip_vs_conn_tab;
 787         ct_read_unlock_bh(idx);
 788
 789         while (++idx < IP_VS_CONN_TAB_SIZE) {
 790                 ct_read_lock_bh(idx);
 791                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
 792                         seq->private = &ip_vs_conn_tab[idx];
 793                         return cp;
 794                 }
 795                 ct_read_unlock_bh(idx);
 796         }
 797         seq->private = NULL;
 798         return NULL;
 799 }
 800
 801 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
 802 {
 803         struct list_head *l = seq->private;
 804
 805         if (l)
 806                 ct_read_unlock_bh(l - ip_vs_conn_tab);
 807 }
 808
 809 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
 810 {
 811
 812         if (v == SEQ_START_TOKEN)
 813                 seq_puts(seq,
 814    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires\n");
 815         else {
 816                 const struct ip_vs_conn *cp = v;
 817
 818 #ifdef CONFIG_IP_VS_IPV6
 819                 if (cp->af == AF_INET6)
 820                         seq_printf(seq,
 821                                 "%-3s " NIP6_FMT " %04X " NIP6_FMT
 822                                 " %04X " NIP6_FMT " %04X %-11s %7lu\n",
 823                                 ip_vs_proto_name(cp->protocol),
 824                                 NIP6(cp->caddr.in6), ntohs(cp->cport),
 825                                 NIP6(cp->vaddr.in6), ntohs(cp->vport),
 826                                 NIP6(cp->daddr.in6), ntohs(cp->dport),
 827                                 ip_vs_state_name(cp->protocol, cp->state),
 828                                 (cp->timer.expires-jiffies)/HZ);
 829                 else
 830 #endif
 831                         seq_printf(seq,
 832                                 "%-3s %08X %04X %08X %04X"
 833                                 " %08X %04X %-11s %7lu\n",
 834                                 ip_vs_proto_name(cp->protocol),
 835                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
 836                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
 837                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
 838                                 ip_vs_state_name(cp->protocol, cp->state),
 839                                 (cp->timer.expires-jiffies)/HZ);
 840         }
 841         return 0;
 842 }
 843
 844 static const struct seq_operations ip_vs_conn_seq_ops = {
 845         .start = ip_vs_conn_seq_start,
 846         .next  = ip_vs_conn_seq_next,
 847         .stop  = ip_vs_conn_seq_stop,
 848         .show  = ip_vs_conn_seq_show,
 849 };
 850
 851 static int ip_vs_conn_open(struct inode *inode, struct file *file)
 852 {
 853         return seq_open(file, &ip_vs_conn_seq_ops);
 854 }
 855
 856 static const struct file_operations ip_vs_conn_fops = {
 857         .owner   = THIS_MODULE,
 858         .open    = ip_vs_conn_open,
 859         .read    = seq_read,
 860         .llseek  = seq_lseek,
 861         .release = seq_release,
 862 };
 863
 864 static const char *ip_vs_origin_name(unsigned flags)
 865 {
 866         if (flags & IP_VS_CONN_F_SYNC)
 867                 return "SYNC";
 868         else
 869                 return "LOCAL";
 870 }
 871
 872 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
 873 {
 874
 875         if (v == SEQ_START_TOKEN)
 876                 seq_puts(seq,
 877    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Origin Expires\n");
 878         else {
 879                 const struct ip_vs_conn *cp = v;
 880
 881 #ifdef CONFIG_IP_VS_IPV6
 882                 if (cp->af == AF_INET6)
 883                         seq_printf(seq,
 884                                 "%-3s " NIP6_FMT " %04X " NIP6_FMT
 885                                 " %04X " NIP6_FMT " %04X %-11s %-6s %7lu\n",
 886                                 ip_vs_proto_name(cp->protocol),
 887                                 NIP6(cp->caddr.in6), ntohs(cp->cport),
 888                                 NIP6(cp->vaddr.in6), ntohs(cp->vport),
 889                                 NIP6(cp->daddr.in6), ntohs(cp->dport),
 890                                 ip_vs_state_name(cp->protocol, cp->state),
 891                                 ip_vs_origin_name(cp->flags),
 892                                 (cp->timer.expires-jiffies)/HZ);
 893                 else
 894 #endif
 895                         seq_printf(seq,
 896                                 "%-3s %08X %04X %08X %04X "
 897                                 "%08X %04X %-11s %-6s %7lu\n",
 898                                 ip_vs_proto_name(cp->protocol),
 899                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
 900                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
 901                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
 902                                 ip_vs_state_name(cp->protocol, cp->state),
 903                                 ip_vs_origin_name(cp->flags),
 904                                 (cp->timer.expires-jiffies)/HZ);
 905         }
 906         return 0;
 907 }
 908
 909 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
 910         .start = ip_vs_conn_seq_start,
 911         .next  = ip_vs_conn_seq_next,
 912         .stop  = ip_vs_conn_seq_stop,
 913         .show  = ip_vs_conn_sync_seq_show,
 914 };
 915
 916 static int ip_vs_conn_sync_open(struct inode *inode, struct file *file)
 917 {
 918         return seq_open(file, &ip_vs_conn_sync_seq_ops);
 919 }
 920
 921 static const struct file_operations ip_vs_conn_sync_fops = {
 922         .owner   = THIS_MODULE,
 923         .open    = ip_vs_conn_sync_open,
 924         .read    = seq_read,
 925         .llseek  = seq_lseek,
 926         .release = seq_release,
 927 };
 928
 929 #endif
 930
 931
 932 /*
 933  *      Randomly drop connection entries before running out of memory
 934  */
 935 static inline int todrop_entry(struct ip_vs_conn *cp)
 936 {
 937         /*
 938          * The drop rate array needs tuning for real environments.
 939          * Called from timer bh only => no locking
 940          */
 941         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
 942         static char todrop_counter[9] = {0};
 943         int i;
 944
 945         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
 946            This will leave enough time for normal connection to get
 947            through. */
 948         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
 949                 return 0;
 950
 951         /* Don't drop the entry if its number of incoming packets is not
 952            located in [0, 8] */
 953         i = atomic_read(&cp->in_pkts);
 954         if (i > 8 || i < 0) return 0;
 955
 956         if (!todrop_rate[i]) return 0;
 957         if (--todrop_counter[i] > 0) return 0;
 958
 959         todrop_counter[i] = todrop_rate[i];
 960         return 1;
 961 }
 962
 963 /* Called from keventd and must protect itself from softirqs */
 964 void ip_vs_random_dropentry(void)
 965 {
 966         int idx;
 967         struct ip_vs_conn *cp;
 968
 969         /*
 970          * Randomly scan 1/32 of the whole table every second
 971          */
 972         for (idx = 0; idx < (IP_VS_CONN_TAB_SIZE>>5); idx++) {
 973                 unsigned hash = net_random() & IP_VS_CONN_TAB_MASK;
 974
 975                 /*
 976                  *  Lock is actually needed in this loop.
 977                  */
 978                 ct_write_lock_bh(hash);
 979
 980                 list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
 981                         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
 982                                 /* connection template */
 983                                 continue;
 984
 985                         if (cp->protocol == IPPROTO_TCP) {
 986                                 switch(cp->state) {
 987                                 case IP_VS_TCP_S_SYN_RECV:
 988                                 case IP_VS_TCP_S_SYNACK:
 989                                         break;
 990
 991                                 case IP_VS_TCP_S_ESTABLISHED:
 992                                         if (todrop_entry(cp))
 993                                                 break;
 994                                         continue;
 995
 996                                 default:
 997                                         continue;
 998                                 }
 999                         } else {
1000                                 if (!todrop_entry(cp))
1001                                         continue;
1002                         }
1003
1004                         IP_VS_DBG(4, "del connection\n");
1005                         ip_vs_conn_expire_now(cp);
1006                         if (cp->control) {
1007                                 IP_VS_DBG(4, "del conn template\n");
1008                                 ip_vs_conn_expire_now(cp->control);
1009                         }
1010                 }
1011                 ct_write_unlock_bh(hash);
1012         }
1013 }
1014
1015
1016 /*
1017  *      Flush all the connection entries in the ip_vs_conn_tab
1018  */
1019 static void ip_vs_conn_flush(void)
1020 {
1021         int idx;
1022         struct ip_vs_conn *cp;
1023
1024   flush_again:
1025         for (idx=0; idx<IP_VS_CONN_TAB_SIZE; idx++) {
1026                 /*
1027                  *  Lock is actually needed in this loop.
1028                  */
1029                 ct_write_lock_bh(idx);
1030
1031                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
1032
1033                         IP_VS_DBG(4, "del connection\n");
1034                         ip_vs_conn_expire_now(cp);
1035                         if (cp->control) {
1036                                 IP_VS_DBG(4, "del conn template\n");
1037                                 ip_vs_conn_expire_now(cp->control);
1038                         }
1039                 }
1040                 ct_write_unlock_bh(idx);
1041         }
1042
1043         /* the counter may be not NULL, because maybe some conn entries
1044            are run by slow timer handler or unhashed but still referred */
1045         if (atomic_read(&ip_vs_conn_count) != 0) {
1046                 schedule();
1047                 goto flush_again;
1048         }
1049 }
1050
1051
1052 int __init ip_vs_conn_init(void)
1053 {
1054         int idx;
1055
1056         /*
1057          * Allocate the connection hash table and initialize its list heads
1058          */
1059         ip_vs_conn_tab = vmalloc(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head));
1060         if (!ip_vs_conn_tab)
1061                 return -ENOMEM;
1062
1063         /* Allocate ip_vs_conn slab cache */
1064         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1065                                               sizeof(struct ip_vs_conn), 0,
1066                                               SLAB_HWCACHE_ALIGN, NULL);
1067         if (!ip_vs_conn_cachep) {
1068                 vfree(ip_vs_conn_tab);
1069                 return -ENOMEM;
1070         }
1071
1072         IP_VS_INFO("Connection hash table configured "
1073                    "(size=%d, memory=%ldKbytes)\n",
1074                    IP_VS_CONN_TAB_SIZE,
1075                    (long)(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head))/1024);
1076         IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1077                   sizeof(struct ip_vs_conn));
1078
1079         for (idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) {
1080                 INIT_LIST_HEAD(&ip_vs_conn_tab[idx]);
1081         }
1082
1083         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
1084                 rwlock_init(&__ip_vs_conntbl_lock_array[idx].l);
1085         }
1086
1087         proc_net_fops_create(&init_net, "ip_vs_conn", 0, &ip_vs_conn_fops);
1088         proc_net_fops_create(&init_net, "ip_vs_conn_sync", 0, &ip_vs_conn_sync_fops);
1089
1090         /* calculate the random value for connection hash */
1091         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1092
1093         return 0;
1094 }
1095
1096
1097 void ip_vs_conn_cleanup(void)
1098 {
1099         /* flush all the connection entries first */
1100         ip_vs_conn_flush();
1101
1102         /* Release the empty cache */
1103         kmem_cache_destroy(ip_vs_conn_cachep);
1104         proc_net_remove(&init_net, "ip_vs_conn");
1105         proc_net_remove(&init_net, "ip_vs_conn_sync");
1106         vfree(ip_vs_conn_tab);
1107 }